ASSET MANAGEMENT

Buy-side AI without the disclosure risk.

Equity research that scales beyond the analyst headcount. ESG screening grounded in real sustainability filings. Marketing review that catches MiFID II and SFDR violations before they ship. All inside one governed AI Gateway.

Customer-hosted on Kubernetes, in your cloud or on premises. For enterprise teams managing AI across business units.

Book a demo

Or email sales@kosmoy.com.

Asset managers were among the first financial services firms to adopt LLMs at scale — research desks were already in the habit of synthesising large volumes of unstructured information. The cycle is now industrialising: front office wants agents on every analyst's desktop, compliance wants a wall around them, and operations is being asked to scale fund commentary and RFP responses without growing headcount.

The regulatory exposure is different from banking but no less sharp. DORA applies in full to AIFs, UCITS managers and management companies. MiFID II governs marketing communications and suitability. SFDR forces sustainability disclosures with auditable underlying data. MAR is unforgiving on the leakage of inside information through any channel, including a shared LLM context window. The EU AI Act adds risk-tier obligations on top.

Kosmoy gives the AM the same operating layer banks use, tuned for the buy-side workflow: research desk segregation, compliance pre-flight on outbound communications, ESG data lineage, and per-fund AI registries that keep Article 8 and Article 9 mandates honest.

What this industry runs into.

Research scaling without context bleed

Equity, credit and macro desks all want their own agents. Without a real isolation boundary, prompts and retrieved documents from one desk end up in another desk's context — a MAR violation waiting to happen.

Marketing and disclosure under SFDR / MiFID II

Outbound material to professional and retail clients can't make performance forecasts, can't claim sustainability outcomes that aren't substantiated, and can't omit principal adverse impacts. LLMs hallucinate exactly the kind of statements compliance can't allow.

ESG data quality

Article 8 and Article 9 funds require that PAI indicators and taxonomy alignment be traceable to source. LLMs synthesise but don't cite reliably — without a governed RAG pipeline they fail the audit trail test.

RFP response factories

Institutional sales receives RFP after RFP. A copilot that drafts answers from past wins is the obvious productivity unlock — but every claim about risk, performance or process needs to be auditable against the source.

Regulatory landscape.

The regulations that shape AI in asset management — and where each one bites on AI deployment.

DORADigital Operational Resilience Act· EU

Applies to UCITS management companies, AIFMs, MiFID firms. Same ICT third-party register and incident reporting obligations as banking.

MiFID IIMarkets in Financial Instruments Directive II· EU

Governs marketing communications, investor categorisation, suitability and best execution. AI-drafted client material is in scope; record-keeping requirements apply to LLM conversations involving order flow.

SFDRSustainable Finance Disclosure Regulation (EU) 2019/2088· EU

Article 6, 8, 9 disclosures require traceable sustainability data. AI-drafted product disclosure statements must cite the underlying source for every PAI indicator and taxonomy alignment claim.

AIFMD / UCITSAlternative Investment Fund Managers Directive / UCITS Directive· EU

Operational and risk management obligations extend to AI used in portfolio management, valuation and risk monitoring.

MARMarket Abuse Regulation (EU) 596/2014· EU

Inside information cannot leak between investment teams. Shared LLM context, vector stores or fine-tuned models can become a MAR breach if research desk segregation is not enforced.

EU AI ActRegulation (EU) 2024/1689· EU

AI for client suitability and recommendation may fall under high-risk uses. Marketing-related AI falls under transparency obligations (Art. 50).

Use cases, and what to measure.

Equity research summarisation

A buy-side analyst covers 35 names across European industrials. Earnings season delivers 20 transcripts in two weeks plus broker notes plus 10-K MD&A sections. The agent reads the earnings call, the prior quarter's call, the consensus model and recent broker notes, and produces a structured note: thesis points, deltas vs consensus, KPIs that moved, management tone. The analyst spends time forming a view, not on extraction. The AI Gateway logs keep the audit trail: which transcript informed which point.

What to measure

Names covered per analyst at the same desk hours, and how often the analyst corrects a thesis point or a consensus delta.

Powered byRAG-in-a-Box Agent Builder AI Gateway

ESG / SFDR due diligence

Portfolio manager queries: 'is XYZ Corp eligible for our Article 9 fund? Check sustainability reports, controversy databases, exclusion lists, and our fund mandate'. The agent retrieves the latest sustainability disclosures, runs the exclusion list, summarises any flagged controversies (with source citations), and produces a recommendation that the PM signs off — not the agent.

What to measure

Analyst hours per new name, and the share of claims in the recommendation that cite a paragraph in a filed document.

Powered byRAG-in-a-Box Action Capsule Guardrails

Marketing material compliance review

Marketing drafts the next factsheet, fund commentary or campaign. The agent runs it against the firm's marketing standards (no past-performance forecasts, balanced risk language, jurisdiction-specific disclaimers, SFDR consistency) and the regulator's published guidance. It flags every passage that is out of policy and suggests a compliant rewrite. Compliance reviewers focus on edge cases.

What to measure

Review backlog in weeks, the share of items the agent flags, and how many of those flags reviewers confirm as real policy issues.

Powered byGuardrails AI Gateway Agent Registry

RFP response drafting

Institutional sales receives a 200-question RFP from a European pension fund. The agent searches past wins for the same firm, similar mandates and similar institutional types, drafts answers, and flags any question where the answer was inconsistent in past responses. Sales and compliance review before submission.

What to measure

Days from RFP receipt to submission, the share of drafted answers that pass sales and compliance review with light edits, and inconsistencies caught against past responses.

Powered byRAG-in-a-Box Agent Builder Insights Dashboard

Portfolio commentary generation

Monthly factsheets need a commentary that reads like a portfolio manager wrote it. The agent reads holdings, performance attribution, sector exposures, and the prior month's commentary, drafts the new one in the firm's house style, and the PM edits/signs. The agent never invents a sector view; it summarises what the data shows.

What to measure

Hours from month-end data to signed commentary across all funds, and how much the PM rewrites before signing.

Powered byAgent Builder AI Gateway Cost Tracking

Agent governance

Where asset management agents need extra discipline.

Asset management agents have to obey two unusual constraints: the wall between research desks (MAR) and the firm-style consistency of outputs (compliance). Kosmoy enforces both as platform features, not policy hopes. Each desk gets its own Agent Registry namespace, its own Action Capsule, its own retrieval set. An equity-desk agent that tries to read credit-desk research fails at the gateway; the Insights Dashboard surfaces the attempt for compliance.

The Agent Registry also captures every fund a given AI system is approved to serve — Article 8 vs Article 9, retail vs professional, EU vs UK distribution. This is the substrate the firm needs to map onto the AI Act, the SFDR product disclosures and the MiFID II target market policy.

Chatbot use cases

Chatbots, by surface and risk class.

Asset managers don't usually ship customer-facing chatbots at the same volume as banks — most distribution is intermediated. The bigger chatbot footprint is internal: research desks, ops, compliance, sales support. Each carries its own governance posture.

Research desk Q&A

Analyst asks 'what was the top-line growth in XYZ's last three quarters and how did it compare with consensus'. Answer cites the source filings — never invents a number.

Operations Q&A

Ops asks 'why is this trade failing settlement?'. Agent reads the break record, exchange messages and the firm's break-resolution playbook, drafts an investigation note.

Sales support

Sales rep asks 'do we have a chart of our European small-cap fund's drawdowns vs index for the last 10 years?'. Agent retrieves the chart from the data warehouse and the standard accompanying language.

Compliance helpdesk

Employee asks 'do I need pre-clearance to buy ABC Corp shares?'. Agent reads the firm's restricted list and personal-account-dealing policy and gives a citation-grounded answer.

How Kosmoy fits.

Asset managers benefit from the same Kosmoy architecture as banks but use different layers more heavily. The AI Inventory becomes the central record of every AI system across funds, mandates and desks. The Gateway carries the policy point — research desk isolation, marketing pre-flight, ESG data lineage. Action Capsules house the agents that interact with the firm's data warehouse, IBOR, accounting book and CRM.

Kosmoy's deployment model — single-tenant, in the firm's Kubernetes cluster — addresses the buy-side concern about IP leakage to LLM providers. Trade ideas and proprietary research never leave the firm's perimeter. The vendor receives zero telemetry from the platform.

Module questions, answered straight.

How do you keep MAR-relevant information from leaking between desks?

Each desk gets its own Action Capsule, its own retrieval index, its own Agent Registry namespace. An equity desk's agent cannot read credit research and vice versa — enforced at the network and gateway layers, not by prompt instruction. Attempts to cross the wall are logged and surfaced to compliance.

Do AI-drafted marketing materials satisfy MiFID II / SFDR?

Kosmoy doesn't sign off on materials — your compliance team does. What Kosmoy does is run a compliance pre-flight against your house standards plus regulatory rule packs, and produce a structured report of every claim that needs review. Compliance signs off on a smaller, focused set of items.

Can we run different agents for different funds?

Yes. The Agent Registry binds each agent to the funds and mandates it serves. An Article 9 fund's portfolio commentary agent has stricter retrieval rules than an Article 6 equivalent, and the AI Inventory keeps that mapping for the audit.

Does Kosmoy support our existing data lake (Snowflake, Databricks)?

Yes. RAG-in-a-Box ingests from Snowflake, Databricks, Weaviate, pgvector and other common stores. The agent retrieves through the Gateway, so retrieval calls are audited and access-controlled the same way as model calls.

Enterprise proof

Built for environments where AI cannot become another unmanaged SaaS silo.

Kosmoy works with regulated and operationally critical organizations, including Banca d’Italia and Leonardo.

Runs in your Kubernetes
Azure · AWS · GCP · on-prem
LLM · MCP · A2A
No mandatory vendor-hosted control plane

Govern the AI on every desk and in every fund.

See how the AI Gateway and Agent Registry handle research desk segregation, ESG lineage, marketing review and per-fund AI mapping.

Book a demo

Or email sales@kosmoy.com.